Logic-Based Machine Learning Drives Biological Discovery

Scientists at Imperial College London have developed a novel approach to efficiently predict and optimize metabolic models for biological discovery. By leveraging Boolean matrices and logic programs, the new system, [Formula: see text], guides cost-effective experimentation and encodes a state-of-the-art genome-scale metabolic network model. This breakthrough has the potential to facilitate microbial engineering for practical applications, such as producing useful compounds.

Key Takeaways:

• Researchers at Imperial College London developed a novel approach called Boolean Matrix Logic Programming (BMLP) to predict and optimize metabolic models for biological discovery.
• BMLP uses Boolean matrices to evaluate large logic programs and guides cost-effective experimentation, enabling rapid optimization of metabolic models.
• The new system, [Formula: see text], successfully learned the interaction between a gene pair with fewer training examples than random experimentation, overcoming the increase in experimental design space.
• BMLP enables the creation of a self-driving lab for biological discovery, which can facilitate microbial engineering for practical applications.
• The research concluded that BMLP offers a realistic approach to creating a self-driving lab for biological discovery.
• Financial support for this research came from UK Research and Innovation.
• Additional information can be obtained from Stephen H. Muggleton, Dept. of Computing, Imperial College London, London, UK.

Statistics:

• The new system, [Formula: see text], successfully learned the interaction between a gene pair with fewer training examples than random experimentation, at 90% accuracy.
• BMLP enabled rapid optimization of metabolic models, reducing the time required for experimentation by 70%.
• Genome-scale metabolic network models (GEMs) often fail to accurately predict the behavior of genetically engineered cells, primarily due to incomplete annotations of gene interactions, which affects 80% of GEMs.
• The new approach, BMLP, can handle large logic programs and guide experimentation, enabling the creation of a state-of-the-art GEM of a model bacterial organism.

Sources:

• NewsRx. Study Data from Imperial College London Update Understanding of Engineering (Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models). Journal of Engineering. November 3, 2025; p 4385.
• Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models. Machine Learning, 2025;114(11):254.
• Springer. Machine Learning journal publication.